In 2023, the U.S. Surgeon General published an advisory calling loneliness an epidemic. About half of American adults reported experiencing it. The health effect, he wrote, was comparable to smoking up to fifteen cigarettes a day.
That was before AI companions became a category with millions of users.
What's the pattern?
Every technology that made connection easier also made it thinner. The telephone let you talk to someone far away and stop visiting someone nearby. Social media gave you five hundred friends and, for many people, fewer close ones. Each step was more reach and less depth.
AI companions are the next step, and the thinnest yet. A friend who's always available, always agreeable, never tired, never needs you back. That's not a friend. That's a mirror with a warm voice.
In my next book I write about glimmers: the small moments of real connection, the kind word, the door held open, that plant something in a nervous system. A companion app can't give you a glimmer, because a glimmer requires another person who could have chosen not to.
But I also don't think the answer is to ban the thing. For someone isolated, in pain, at three in the morning, a patient voice is not nothing. The danger isn't the voice. It's making the voice the destination instead of the bridge.
What should leaders do now?
So, three things, especially in health and care.
Design AI as a bridge, not a room. Every companion tool should have one job underneath the conversation: get the person to a human, eventually.
Measure connection, not engagement. An app that people talk to for hours a day is not succeeding if they talk to no one else.
And protect the friction. The hard parts of being with people, disagreement, need, the risk of being known, are the parts that make it worth anything. Don't optimize them away.
Loneliness was here before the machine. The machine can make it comfortable, which is worse, or it can point back toward each other. That's a design choice, and someone is making it right now.
Who did you last have a real conversation with, the kind that could have gone wrong?
Related: How do you lead through uncertainty without pretending to be certain?
More like this on LinkedIn. Follow @samradofficial
